SES AI Corporation (SES) Earnings Call Transcript & Summary

Welcome to today's fireside chat with SES AI. I'm your host, Brian Gordon, Senior Research Analyst at Water Tower Research, and I'm joined by Shawn Severson, CEO and Co-Founder of Water Tower. Today, we are joined by Qichao Hu, Founder, Chairman and CEO of SES AI. Welcome, Qichao. Welcome Qichao.

Thank you, Brian.

As investors may be familiar, SES AI is revolutionizing electric transport and other emerging industries and market with its pioneering next-generation lithium metal batteries. The company is a leader in AI for materials discovery using leading-edge AI to augment its R&D. The company is also a leader in AI for battery manufacturing and for battery safety. There's a lot of ground to cover. So I was thinking that we just get started, Qichao.

My first question, why don't we start with a bit of a history and an overview of the company in the emerging markets it serves, which include EVs, UAMs, urban air mobility, drones, robotics and grid and data centers. Maybe start out with a little bit about how the company started and tell us about the primary markets you guys are targeting.

Yes. So we started focusing on lithium metal batteries. And then the key to making lithium metal battery work has always been electrolyte development. So since 2012, for the past 12 years, we've been focused almost exclusively on electrolyte development. And so we worked on electric lithium metal batteries for drones applications and the EV applications. And then recently, we found that because the key is in the electrolyte, and then we really want to go one step beyond, and that is to map all the possible materials that could be used for battery electrolyte. And this is not a new idea itself. It's an old idea, but then it was never possible before with the old hardware and the old computing hardware and software. But with the latest GPUs and then GPU accelerated computation chemistry software, we can actually map all the small molecules, all the physicochemical properties of the small molecules. And so we did that. And then recently, we expanded to electrolyte for any lithium batteries, not just lithium metal, but also lithium ion. So we expanded to beyond just lithium metal for EVs, which we did with Hyundai, with Honda, with GM. And now we are expanding to also lithium ion for robots, drones and energy storage. And then we found that this tool itself is actually really powerful. And then we can use AI not only to compute for the physical chemical properties of the molecules, and we can actually use that to help us screen and then find the interesting molecules. And so recently, we announced this new battery 2170 is actually a lithium ion battery. It's not lithium metal, it's a lithium-ion battery. But what's unique about it is the electrolyte. So that really helped us going from just a lithium metal to EV storage to now lithium metal and lithium ion for drones, robots, EV and stationary storage, a much bigger market.

So that's really interesting. I think maybe it would be worthwhile to kind of step back just a little bit and talk through why different battery chemistries are important for these different applications?

Yes. So different applications have different needs for cost, cycle life, safety, energy density, power density, charge rate, discharge rate. It's just different applications. For example, the unmanned applications like drones, robotics, they care less about safety, more about power density, energy density. And then the main applications like EV, UAM, care mobile safety, less about power density. Stationary storage care more about longevity, cycle life, less about power density. So the different needs require different chemistry. But all these lithium-based chemistries have one thing in common, and that is lithium plating. So in the case of lithium metal, you get lithium palated in every cycle. That's what you do because you do not form a stable SEI, so you always plate lithium. In the case of lithium ion, you also have lithium plating. It's not every cycle. It happens when you do low temperature, fast charge and also after long cycles. So -- but that problem does also exist. For example, LFP for stationary storage after long cycles. Silicon-based anodes when you do fast charge and then low temperature. And we found that the electrolyzed solvent that we did discover through mapping the molecular universe can address lithium plating for both lithium metal and lithium-ion. That's why the application -- even though different chemistries for different applications, electrolyte solvent can be used across the board.

And why is plating problematic for battery, just like in the big picture?

Yes. So we have -- so plating itself is okay. It's a nonuniform plating. So if you play uniformly and densely, that's okay. But when you have nonuniform and then local plating, then you accumulate massy lithium, And eventually to dendrite. So that could either short the battery or in the subsequent cycles, your lithium, your current distribution just is not uniform. So that caused degradation in the battery performance.

So I definitely want to get back to AI and how you guys are using AI because I think what you're doing, combining both the modeling and the empirical work with the electrolyte foundry is really interesting. But I wanted to kind of have you walk a little bit through sort of the bigger picture of the battery industry. As investors, I'm sure, are very familiar, the industry today is dominated by several giants, companies from China, from South Korea, from Japan. Could you talk about how SES AI is situated within the industry? And especially given how you see the industry evolving with these new chemistries, with these new applications coming online. How do you think that the industry itself is going to evolve and your position within the industry over the next, let's call it, 5 years or so?

Yes. I think in terms of the large -- I think there's 2 aspects. One is the large established manufacturing and then another is the emerging markets. For example, the large established consumer markets, EV and then stationary storage, I think these currently are dominated by Asian players. And it's really hard for a U.S. company to compete in terms of manufacturing batteries at that scale at their cost. But we also don't have to because the cause of a battery, I mean now in LFP lithium-ion battery is like $35 per kilowatt hour, right? So it's really low cost. But it's a benefit for the development of AI because that just means your data acquisition as well. It's not so good as a battery company, but it's really good as an AI company. And then another aspect is the emergence of new markets like drones, robotics for defense applications. And then for all sorts of geopolitical reasons, Asian companies cannot really participate in that market because of defense, all that. But then these are markets that will take off drones, drones warfare will be the future warfare. And then -- and I think U.S. companies, including SES, will have an advantage there in terms of not just developing the battery technology, but also manufacture there. So I think both for the consumer market, we as an AI company, the cost of data acquisition is low. And then also in the emerging drones and robotics, the market is also high as a battery company.

Qichao, I have a question to just jump in. Since the beginning of the business and covered you for quite a while, you've always talked about the manufacturability of batteries as well. So I think investors hear a lot about great stuff being done in the lab, right? There's really interesting things done in solid state, but they don't have the ability to scale to commercial volumes, all kinds of manufacturing issues. And I found part of your story being very unique that when you think about technology, you think about innovation, how maturing -- manufacturability and design for manufacturability comes into play. Can you tie that back to how you're using AI, what you're doing your approach overall to technology development and commercialization?

Yes. So from the beginning, we never wanted to change the manufacturing process. So for the consumer side, EV and stationary storage, if you think about what we do as an AI company, we don't change the manufacturing process. We follow the manufacturing quality control process using AI for manufacturing to ensure quality and is defect free. And then we also collect the data for example, that 2170 battery that we use, the AI-Enhanced 2170, that does not change anything in the manufacturing process. All it changes it just electrolyte. So before it used to fill with a different liquid electrolyte, now you fill with a new liquid electrolyte. Actually, we have a contract manufacturer that builds the dry cell and then we just fill in our electrolyte. So that part, we don't change the manufacturing process. And for the emerging drones and robotics, we also don't change the manufacturing process. It's the same as lithium-ion, the pouch cylindrical, we do not change that. And in the anode, we changed the material, the anode, we use lithium metal. We also now use high silicon for the AI-Enhanced, but we never changed the manufacturing process. I think that has always been the key because one of the most important things about battery is you have to get to that scale. EV, stationary, you obviously need scale and then drones, robotics, warfare, no one fights a war with like 2 drones, right? You fight with millions of these. So the scale is -- has been quite important for us from the beginning.

And that enables you to use existing manufacturing technologies, existing infrastructure, and that translate all that into good cost competitiveness if I look at it that way, being able to leverage the existing manufacturing infrastructure and knowledge base out there is key to this because it creates lower cost commercialization, right?

Yes.

So I definitely want to dive into how you guys are using AI for Discovery because I think what you guys are doing is a little bit unique. You are using large language models and you're using it to help mine the scientific literature. But you're also using foundation models that are built specifically for the chemistry using like graphical neural networks. And then in addition to those 2 kinds of models, you also have your electrolyte foundry. Could you talk a little bit about what each of these types of AI for discovery brings and how, in particular, having that empirical side, having that electrolyte foundry then feeds back into what you can do on the modeling side?

Yes. So in terms of AI, there are a few steps. One is, first, we need to generate the data. And then a lot of the molecule level material data currently do not exist. So you have to have that data to feed these models. And then if the data don't exist, we have to generate the data. So we start with computation chemistry, not even AI, just high-performance computing and then we compute, for example, the single molecule energy levels, the solution level solubility, melting point, balling point and then the interface, the properties. First, we compute all these different properties on 10 to the 8th and then 10 to the 9th and eventually 10 to the 11th, all the small molecules. Why small molecules? Because all the electrolyte, the solvent cell additives, all 3 components are made of small molecules, less than 20 atoms. So we synthesize the data first. And then -- okay, so you have the property data on 10 to the 11th. Obviously, it's not feasible to synthesize to actually go to the actual foundry to make all these molecules and then test them in the batteries. Realistically, you can probably make 1,000 a year, test them in the lab. So how do you go from 10 to the 11th, 100 billion to 1,000. So this screening process require model development. And then this part, we have -- the benefit that we have is we have a team of really good human scientists. So we train these models with all the papers that we can get access to. I think we are up to like 90 million papers, basically all the battery-related literature, all the organic chemistry related, all the material science-related papers, we feed the model and then the [ box ]. So it has a good understanding, but still not good enough. And then we developed this agent, and then we have a team of human scientists that basically just -- so one, annotate all the papers because you can't just feed the papers. And then you do have to annotate the papers and then the human scientists will interact with the agent and then ask a question. For example, what is the interfacial reaction between this type of solvent on this type of anode. What should be the composition of the SEI. And then the agent will provide a response. The human scientists will provide feedback, okay, this is wrong, this is right, and then it should be this, back and forth a lot of times. And then finally, you will have a super intelligent agent that will tell you how to filter from 10 to the 11th down to 1,000. So -- and then we make about 100 molecules per month, so a year, about 1,000. And then we go to the foundry and actually synthesize these solvent molecules. All these are new. And then we formulate them in electrify and actually test them in batteries. And then also along the way, for example, when we compute the properties like melting point and the bulling point and solubility, sometimes we are off. And then sometimes we don't know if that's accurate. So then we actually have to make the molecule and then actually test the solubility, actually test the melting point and actually provide that feedback. And then we also work with other research labs try to collect and in some cases, buy these data so that we have the real data and then we can use the real data to train the computation models and then make these adjustments.

That's what I personally think is the most interesting about the approach that you guys have. Because over time, what that allows you to do is create a very proprietary set of data that will really sort of augment your discovery and your ability to actually accelerate the commercialization process.

Yes. Yes. And this tool, this model is quite powerful. And we have scientists who are really experienced like 15 years, 20 years plus experience in the field. But I mean -- and then if you ask them what they go through in terms of thinking process when they have to develop a new electrolyte, they also don't really know. They have these high level of things, but they don't really know. For example, in silicon batteries, a couple of years ago, the industry developed FEC. FEC as a solvent was developed in particularly for silicon based anodes. It wasn't for graphite based anodes. So how was that discovered. And so next, how can we discover another new solvent that's similar to FEC, but bypass the human thinking process and then rely on this agent.

Now, I think on your latest investor presentation, you talked about something like 17 or 18 new potential electrolytes that you've identified and that cure in production, does that include the new 2170 cell? Or is that sort of like an incremental discovery above what you guys have disclosed previously?

Yes. So that was the old one. That was after we mapped, I think, 10 to the 5th molecules, and we identified 17 or so new molecules. But since then, we are now on to 10 to the 9th, and we have a lot more need to synthesize and then verify.

And is this like over the next year or 2 years? Like how long will actually take you given the throughput of the boundary to map those kind of empirically?

Yes. So mapping the molecular universe, just in terms of computing the properties, that's pretty fast. The single molecules like 2 months and then the whole block of universe is done. At the solution level, that will take longer. But we expect this year 2025, we can map the solution -- not just a single molecule level, but given the solution level properties of both organic and inorganic molecules. And then in parallel, come out to about 100 molecules synthesis per month.